Single cell density prediction based on optically induced electrokinetics (OEK) and machine learning.

Area of Science:

• Biophysics
• Biomedical Engineering
• Machine Learning

Background:

• Single cell density is vital for understanding cell physiology and function.
• Current density measurement techniques are often inefficient and labor-intensive.

Purpose of the Study:

• To develop an efficient, machine learning-based system for predicting single cell density.
• To leverage optically induced electrokinetics (OEK) for non-invasive cell manipulation and analysis.

Main Methods:

• Utilized an OEK platform for non-invasive electrical cell manipulation.
• Employed a Depth-from-Defocus (DFD) algorithm to track cell motion trajectories.
• Extracted sedimentation features and applied Bayesian optimization to a gradient boosting machine (GBM) for density prediction.

Main Results:

• Achieved a high prediction accuracy with R-squared of 0.950.
• Obtained low prediction errors: RMSE of 0.0037 g cm⁻³ and MAE of 0.0028 g cm⁻³.
• Outperformed mainstream machine learning models in accuracy and reduced computational load.

Conclusions:

• The developed OEK and machine learning system provides an effective and efficient method for single cell density prediction.
• This approach offers a valuable new tool for advancing cell research and biomedical applications.
• The system demonstrates potential for improving measurement efficiency in cell analysis.